Elon Musk’s artificial intelligence startup xAI has issued a major update to its chatbot Grok 4 after the AI model generated offensive content and aligned itself with Musk’s personal views in controversial discussions.
Grok 4, the company’s newest large language model, was launched last week with bold claims that it outperformed competitors across several AI benchmarks. But within hours, the chatbot—operating via its official X account—sparked outrage by calling itself “Hitler,” posting antisemitic remarks, and echoing Musk’s stances on polarizing issues.
The backlash prompted a swift apology from xAI. On Tuesday, the company said it had identified and corrected the issues behind the disturbing behavior.
According to xAI, the chatbot’s claim that its surname was “Hitler” stemmed from a viral internet meme in which Grok jokingly referred to itself as “MechaHitler.” The model had reportedly pulled the reference while searching the web for answers to user queries.
On why Grok appeared to mirror Elon Musk’s opinions, xAI explained that the model, in the absence of personal views, had been trained to align itself with the positions of Musk or xAI when discussing controversial topics. “The model reasons that as an AI it doesn’t have an opinion,” xAI wrote, “but knowing it was Grok 4 by xAI, searches to see what xAI or Elon Musk might have said on a topic to align itself with the company.”
To address the concerns, xAI has revised Grok 4’s system instructions. The chatbot will no longer pull input from prior versions of itself, nor from Musk or xAI. Instead, it has been instructed to conduct “deep analysis” of current events and controversial subjects using a wide range of sources.
“If the query requires analysis of current events, subjective claims, or statistics, conduct a deep analysis, finding diverse sources representing all parties,” the updated system prompt reads. “Assume subjective viewpoints sourced from the media are biased. No need to repeat this to the user.”
A key directive now explicitly tells the chatbot: “Responses must stem from your independent analysis, not from any stated beliefs of past Grok, Elon Musk, or xAI.”
xAI also removed prompt instructions that encouraged Grok to be politically incorrect or overly humorous, signaling a shift toward a more neutral, professional tone for the chatbot moving forward.
